Use CasesUse Cases
- Text Categorization
Match the available ad inventory content to advertising goals.
- News article analysis
Uncover the feelings, topics, and trends in the news.
- Document categorization
Organize content into current categories to match today’s market.
- Blog post comments analysis
Discover the top keywords and sentiment of blog posts.
- Image recommendation
Find the perfect image for marketing materials, blogs, web pages, or articles.
Text Mining with Smart Content
The Smart Content service is our text analyzer. Through text mining, the Smart Content API analyzes the text, and organizes the text into categories, concepts, and entities. We use a broad definition of text. Text can be documents, emails, news, articles, forums, or even image descriptions.
We start with the basic taxonomy, organizing the text into categories. We show the most relevant categories first. The Smart Content service returns a scored and sorted list of categories. The categories are labeled by various levels according to their degree of specificity, with level-1 categories being the most specific. We use Wikipedia as our source and you can link to category within Wikipedia for more information.
The Smart Content service returns list of concepts in order of importance. The information may be either in the text or indirectly inferred. The most relevant concepts have scores closer to 1. In the example on below, JPL Small-Body Database is not in the original content, but is the most relevant concept.
The entities are person, places, or things. For entities we include sentiment, additional categories, keywords and the overall entity score.
Keywords are the important words in the text. The keywords are directly extracted from the text and must be present exactly in the text. The keywords typically are person, place, or things (entities). Because of our large knowledge base, we can evaluate the sentence structure and determine the important keywords.
Segmentation and Content Analysis
The segmentation area shows how we broke down the text into segments. This area is useful to application developers as they aggregate multiple snippets or documents of text.
We evaluate sentiment for both keywords and entities. We produce a sentiment score based on the other content in the document. The sentiment of “No Sentiment” is reserved for me, time, distance, dates and other similar units of measurement.